Continuing my discussion of Rashid's question from the last issue, having to do with step-response testing of serial links, I'll show how you can squeeze a lot of information out of that one simple test.
Confirm the Diagnosis
Today we will measure signals on a high-speed 3.125 Gb/s differential channel. The signal emanates from a National Semiconductor DS25BR100 LVDS-style differential driver, mounted on an evaluation board. The driver incorporates transmitter pre-emphasis, a feature that is turned OFF for these tests.
As depicted in Figure 1, the driver outputs (TX+ and TX-) connect first to a closely-spaced, tightly-coupled, 100-ohm differential pair. The trace spacing is dictated by the pin spacing on the driver package. The tightly-coupled pair runs for a distance of 0.5 inches away from the driver. After that short run the traces separate, making two independent, non-coupled, 50-ohm transmission lines. The purpose of separating the pair is so you can use fatter traces, reducing the skin-effect loss. The separate traces each run 1.7 inches to the SMA output connectors, which continue on 50-ohm semi-rigid low-loss coaxial cables to a distant receiver.
This is the same board I used in response to Rashid's question in the last newsletter, "Step Response Test".
To begin, let's observe the signal at the points indicated by dark purple arrows (A) in the layout drawing, Figure 2, using a LeCroy D600A-AT differential probe. Right now the system is transmitting a very slow, repetitive square-wave pattern.
According to the measurement, the 100-ps step edge looks great, but is followed 400 ps later by a nasty negative bump. That bump represents a form of intersymbol interference that can, if the next signal edge occurs 400 ps later, create jitter. (see: "Step Response Test".)
My preliminary conclusion, looking at this problem, is that the negative bump probably comes from the SMA connectors. What I'd like to explain to you is how I arrive at that conclusion, and what I do next.
A time-space diagram illustrates the life history of one and only one step edge as it makes its way from the driver through the system, bouncing along the way off various obstacles, until all the pieces of the signal dissipate to negligible levels.
Near the top of the diagram I depict the system in a simple linear fashion. The driver (source-terminated) appears on the left. The driven signal proceeds from left to right. The actual system is a differential layout; only one leg of the differential system appears in this simplified schematic view.
The vertical dimension of the chart represents time, proceeding downward. The chart depicts two independent scenarios.
Scenario #1 (blue) begins at time t, with a single rising edge from the driver. The blue diagonal streak depicts the progress of that edge in both time and space as it makes its way towards the right side of the diagram.
Scenario #2 (in red, completely unrelated to the first) begins at time t.
I hope you can see from the diagram that these two generic sequences are the only two possible ways to generate a main step edge followed by a bump. Either the signal rattles around before it arrives at the probe (#1), or it passes the probe first, then reflects back (#2). In either case the reflection must be a reactive one, generating from each step edge a short bump.
So, if there are two possibilities, why do I conclude that the problem is likely happening at the SMA connector? There are three big hints.
First, I count the reflections. If the bouncing happens to the left of the probe (#1), the reflected bump must bounce at least 2 times before it arrives at the probe. The situation on the right side (#2) differs, because the bump arrives as a backward-moving signal. It bounces only once.
If, as tends to happen in high-quality both-ends terminated systems, the signals shrink markedly at every reflection, then reflection problems to the left of the probe will naturally tend to create smaller objectionable features than equivalent-sized reflections occurring to the right of the probe. Put another way, I look for the simplest explanation first—one reflection is simpler than two.
Second, I checked the timing. In the previous newsletter I found that the total round trip time from driver to probe was, because of the limited distance, only 278 ps, not enough to account for a 400-ps delay. On the right side, between probe and SMA connector, my estimated round trip delay was 457 ps, a much closer match to the observed timing.
Third, I know from experience that connector layouts often cause problems.
Now let's get down to business. I need to confirm my judgment with additional measurements. The confirmation step is crucial because it takes a lot of time to do re-work, or re-layout, and you must be sure of your conclusions.
The first thing to check is the probe. You know the input impedance of a probe, especially at very high frequencies, affects every circuit it touches. If the probe itself somehow causes the reflected bump, I'd like to know right away.
How can you test that? Easy. Use a second probe.
Touch the second probe onto the circuit right next to the first one. While you do that, monitor the signal using the first probe. When you touch, an enlarged bump implicates the probe. A same-sized bump clears the probe of blame.
So, is my probe a problem? I already gave you the answer to that question. Figure 2 shows the response when measured with a single probe, and measured the same way but with a second probe also touching the circuit at the same time. I offset the waveforms vertically for visual inspection. The bump appears the same. Probes exonerated.
I could have told you that without doing the measurement, because I know (from looking at the probe specs) that the differential input impedance of a LeCroy D600A-AT 7.5 GHz differential probe at 1.25 GHz is about 1600 ohms. That value is so high that it should make, in this setup, reflections no greater than 3%. The negative bump appears huge compared to 3%, so it could not have come from the probe.
Next, let's test my timing calculations. If the bump really happens due to the round-trip delay action t-t-t, then if I move my probe to a location physically closer to the SMA connector the bump should then occur at a time closer to the main signal edge. Figure 4 shows measurements made at the original position (A), plus new positions (B) and (C) marked in Figure 1. In this figure the scope triggers on a sync output from the signal source that feeds the DS25BR100 driver, so that at each successive point of measurement you can see the additional delay on the main signal edge. Checking the main edge delay confirms for me a trace propagation delay of 165-172 ps/in., depending which measurement you take. These numbers are all in line with my earlier blanket assumption about trace delay. No surprise here.
Now I line up the times of arrival for all the signals so I can see the effective delay, from edge to negative bump, for each case. I like to do this in MathCad. You could use MatLab just as easily. The LeCroy scope stores waveform files directly in MathCad, MatLab, Excel, and other formats. A good math spreadsheet gives you complete control over the final format of the display, composition of multiple waveforms, calculation of eye patterns, and other technical issues necessary to good-quality lab documentation.
The zoom-view picture I inserted at the bottom right of Figure 4 clearly shows a regular progression of bump timings. As you move the probe closer and closer to the SMA connector, the bump move progressively closer to the main signal edge.
Given the way the bump moves, I'm now certain it comes from the SMA location—and I'm equally certain that it has no effect on final system operation. That's because the bump is observable only within the confines of the demo board. At the board boundary (the SMA location t in Figure 3), the bump is initially directed back into the board, from whence it must make another reflection before emerging. In the process of bouncing, I assume it becomes much smaller.
You can check that assumption by looking at any of the waveforms in Figure 4. If some double-reflected bump were to emerge after the first negative bump, you would see it passing by. Yet, such a feature does not appear. From this clue I deduce that the main signal exiting the SMA connectors probably looks fine. That is a common occurrence in both-ends terminated systems (see "Both-Ends Termination").
Apparently, we've been chasing a problem that isn't really a problem, and that brings me to my last point: if you want to check signal quality, do so at the bitter end of your transmission structure, where it matters.
I hope that through this discussion you have learned some useful measurement techniques. Checking your measurements by doubling up the probes, and changing the probe location, can help improve the accuracy and reliability of your final conclusions.
I'll talk a lot more about measurements and their role in the debugging process in my all-new seminar,
High-Speed Noise and Grounding.
April 30-May 1, 2008 in Washington, DC,
—right after I deliver my ever-popular introductory course,
High-Speed Digital Design.
April 28-29, 2008 in Washington, DC,
A direct results of collaborations with many groups over the past years working on mixed-signal systems, the new course addresses critical issues of noise and grounding in advanced applications including wireless communication, cell phones, avionics, GPS devices, telemetry, and guidance systems. That course will help you identify and solve crosstalk problems within any mixed-signal system.
Dr. Howard Johnson
Thanks to LeCroy for the use of their SDA 6020 scope, which made these measurements possible. Thanks also to National Semiconductor for the use of their evaluation board.
Take Your Medicine
When I was a kid, each time after I got into trouble my parents forced me to make an apology. It was a bitter experience, something they called "taking your medicine". Now that I'm older and have a more philosophical view of life, the formal apology has evolved into a great way to unload some of those little burdens we all carry. I'm going to unload one right now.
To Lee Sledjeski, at National Semiconductor, who noticed while reading a pre-production copy of this article that the reflected bump from the SMA connector on his board was not as bad as mine, I hereby confess that I artificially enlarged the bump to make it appear sharper and clearer.
The board as designed exhibits only a small bump (in blue) from the SMA connector. My board shows a larger bump (in red).
In order to make a more perfect article, and with the hope that you will understand the imperative of having good examples for pedagogical purposes, I used an old microwave engineers' trick called a "tweak". It's a small bit of copper, roughly 0.2 in. on a side, usually triangular, soldered to the SMA connector right at its signal pad. A tweak sticks straight up into the air. It increases the parasitic capacitance of the SMA signal pin, enlarging the negative bump. It makes no observable difference to the final output waveform, for reasons explained in the article.
To my friends at National Semiconductor, I apologize for abusing your board in this way. The tweak was not used with any intention of making the board look bad. In fact, the board is so robust it works fine even with the tweak in place (Figure 5).
Perhaps I can explain my actions by pointing out the utility of tweaks, and the desirable benefits of teaching my readers about them. A small tweak, glued to the end of a wooden stick, instantly makes a beautifully crisp reflection wherever, and whenever you desire. For example, if you want to add some ISI, or some jitter, to a waveform, a strategically placed tweak can do the job. Touch it to various locations in your circuit to check the correspondence between reflections visible in your step response and physical locations in the layout. A tweak placed at the SMA location in this article, for example, would have instantly confirmed the physical location causing the negative bump.
Now that you know about tweaks, you may find yourself using one someday to delay a signal by a few tens or hundreds of picoseconds for the purpose of fixing (temporarily, at least), a timing problem.
I first learned about tweaks from Dave LaCombe at Omega Microwave in Sunnyvale, CA., in about 1987. He used them to adjust the frequency response of delicate microwave circuitry. Once soldered in place, Dave would turn or bend the tweak until he obtained a satisfactory response.
Thanks, Dave, for getting me into trouble.